Fiduccia Ventures Builds an Agentic AI Recruitment Pipeline on AWS for Consistent, Evidence-Based Hiring
RECRUITMENT & TALENT SOLUTIONS · AWS GENERATIVE AI
Project Overview
Fiduccia Ventures is a recruitment and talent-solutions organization supporting recruiters, hiring managers, and enterprise talent-acquisition teams across the full hiring lifecycle — from role definition and candidate screening through structured interviewing to the final hiring decision. Its platform serves five core personas — recruiters, hiring managers, interviewers, candidates, and HR administrators — through a recruiter portal, a mobile app, WebRTC video interviews, email and notifications, and API/ATS integrations.
To modernize this process, Fiduccia Ventures engaged ILIOS Digital, an AWS Generative AI Competency Partner specializing in agentic AI/ML platforms for enterprise and public-sector clients, to design and implement an agentic, AI-powered recruitment pipeline on AWS. The solution chains together Amazon Bedrock Agents to understand hiring intent, generate interview plans dynamically, run real-time evidence-based interviews with adaptive follow-up questioning, and produce a structured, recruiter-ready assessment — while keeping a human recruiter firmly in the loop for the final decision.
Challenges
Interview quality and consistency at Fiduccia Ventures had historically depended on the individual recruiter or interviewer. Writing role-specific questions, keeping a live conversation aligned to job requirements, probing follow-up areas when an answer was incomplete or unusually strong, and pulling together evidence from resumes, applications, pre-screen answers, video and audio, and chat transcripts into a defensible recommendation — all of it was manual, judgment-driven, and slow.
Heavy Manual Prep
Recruiters and hiring managers spent significant time manually writing role-specific interview questions and job descriptions for every open requisition.
Scripts That Couldn't Adapt
Static, pre-written scripts couldn't adapt in real time, so interviewers missed chances to probe promising or concerning answers with targeted follow-up questions.
Slow, Inconsistent Evidence Checks
Evidence from resumes, applications, pre-screen answers, video/audio, and chat had to be manually cross-checked against job requirements and competency rubrics — slow and inconsistent across interviewers.
Assessment as a Bottleneck
Consolidating interview evidence into a structured, comparable assessment was a manual step that delayed decisions and slowed time-to-hire.
Rigor That Varied by Interviewer
Interview rigor varied by interviewer, with no systematic way to align live questioning to company policy and evaluation rubrics.
Integration Constraints
Any new platform had to integrate with existing recruiter portals, a mobile app, WebRTC video tooling, and external ATS/API clients without disrupting established workflows, while meeting enterprise-grade security and disaster-recovery requirements.
Solutions
ILIOS Digital designed an agentic, event-driven recruitment orchestration platform on AWS. Amazon Bedrock and Amazon Bedrock Agents interpret hiring intent and context, dynamically generate interview plans, run real-time interviews with adaptive follow-up questioning, evaluate evidence against job requirements, and synthesize a structured assessment for recruiter review.
Five-Stage Recruitment Flow
Intent & Plan Generation
A recruiter or hiring manager opens a requisition. The orchestration agent interprets intent and generates a role-specific interview plan grounded in role libraries, competency models, and company policy.
Real-Time Interview Orchestration
During the live interview, the orchestrator manages the conversation while Amazon Transcribe converts audio to text and Amazon Comprehend extracts entities, sentiment, and key phrases in near real time.
Evidence Evaluation & Adaptive Probing
The evidence evaluator agent continuously checks responses against job requirements and autonomously triggers targeted follow-up questions where evidence is incomplete or noteworthy.
Assessment Synthesis
Once the interview ends, the assessment synthesizer consolidates all evidence into a structured hiring assessment, which AWS Step Functions and AWS Lambda finalize into a PDF/dashboard-ready report.
Human Decision
The finished assessment hands off to the recruiter, who makes the final call.
Platform Architecture
Underneath the agent chain, the platform runs on five cooperating layers: a data and intelligence layer (Amazon S3 data lake, Amazon Transcribe, Amazon Comprehend, Amazon DynamoDB, Amazon Bedrock reasoning); a knowledge store of job descriptions, competency models, and policies indexed for vector-based semantic search; a workflow layer using AWS Step Functions, AWS Lambda, and Amazon EventBridge; a delivery layer fronting the recruiter portal, mobile app, video interviews, and ATS/API clients through Amazon API Gateway; and a security and monitoring layer spanning AWS IAM, AWS KMS, Secrets Manager, WAF, Config, CloudWatch, X-Ray, and CloudTrail.
The platform runs across a primary region with a secondary disaster-recovery region, using S3 cross-region replication, DynamoDB Global Tables, EventBridge Global Endpoints, and RDS Multi-AZ, with Route 53 health-check-based failover and AWS Backup for scheduled recovery.
Platform Layers
- ✓Data & Intelligence Layer: S3 data lake, Transcribe, Comprehend, DynamoDB, Bedrock reasoning
- ✓Knowledge Store: Job descriptions, competency models, policies with vector-based semantic search
- ✓Workflow Layer: Step Functions, Lambda, EventBridge for event-driven orchestration
- ✓Delivery Layer: API Gateway fronting recruiter portal, mobile app, video interviews, ATS/API clients
- ✓Security & Monitoring: IAM, KMS, Secrets Manager, WAF, Config, CloudWatch, X-Ray, CloudTrail
- ✓Resilience: Multi-region with cross-region replication, Global Tables, failover, and managed backup
AWS Services
Intent & Interview-Plan Generation
Services: Amazon Bedrock Agents (Orchestration & Plan Generator)
Automates contextual job descriptions and role-specific interview questions and flow.
Real-Time Adaptive Interviewing
Services: Amazon Bedrock Agents (Real-Time Interview Orchestrator)
Dynamically selects follow-up questions based on candidate responses as the interview happens.
Speech-to-Text Transcription
Services: Amazon Transcribe
Converts interview audio into accurate, searchable text for downstream evaluation.
Entity, Sentiment & Key-Phrase Extraction
Services: Amazon Comprehend
Surfaces structured signals from transcripts and text responses to support evidence scoring.
Evidence Evaluation & Assessment Synthesis
Services: Amazon Bedrock (foundation models)
Cross-checks evidence against job requirements and generates a structured hiring recommendation.
Candidate & Session Data Store
Services: Amazon DynamoDB
Maintains candidate profiles, interview history, scores, and live session state.
Workflow Orchestration
Services: AWS Step Functions
Coordinates the end-to-end interview-to-assessment workflow across all agents.
Event-Driven Compute & Integration
Services: AWS Lambda, Amazon EventBridge
Executes business logic, scheduling, triggers, and notifications cost-efficiently at scale.
Data Lake & Storage
Services: Amazon S3
Stores raw interview documents, audio, transcripts, evidence packages, and knowledge-base content.
Access & Delivery
Services: Amazon API Gateway (REST/WebSocket)
Fronts the recruiter portal, mobile app, video-interview, and API/ATS channels.
Security & Governance
Services: AWS IAM, AWS KMS, Secrets Manager, AWS WAF, AWS Config
Enforces access control, encryption, secrets management, and configuration compliance.
Monitoring & Observability
Services: Amazon CloudWatch, AWS X-Ray, AWS CloudTrail
Provides metrics, distributed tracing, and API audit logging across the platform.
Disaster Recovery
Services: Amazon Route 53, AWS Backup, DynamoDB Global Tables, S3 Cross-Region Replication
Enables cross-region failover and scheduled backup/restore for business continuity.
Return on Investment
Fiduccia Ventures' Expected Gains
Interview-Question Prep Time per Requisition
BEFORE
Manual drafting for every role
AFTER (TARGET)
Auto-generated plan in minutes
ROI / QUANTIFIED VALUE
Prep reduced to minutes
Follow-Up Question Relevance
BEFORE
Dependent on interviewer skill; static scripts
AFTER (TARGET)
Dynamically targeted to each candidate's responses
ROI / QUANTIFIED VALUE
Real-time adaptive probing replaces static scripts; relevance uplift to be measured
Evidence-to-Assessment Turnaround
BEFORE
Manual consolidation, 3 days
AFTER (TARGET)
Generated immediately after the interview, 2 hours
ROI / QUANTIFIED VALUE
3 business days → 2 hours = 91.7% reduction in elapsed working-hour turnaround
Scoring Consistency Across Interviewers
BEFORE
Not systematically tracked; manual, subjective
AFTER (TARGET)
Scored against a common rubric by the evaluator agent
ROI / QUANTIFIED VALUE
Common rubric + evaluator agent; consistency improvement to be measured against baseline
Recruiter Manual Workload per Hire
BEFORE
High — scripting, note-taking, manual synthesis
AFTER (TARGET)
Reduced to review and final decision on a structured assessment
ROI / QUANTIFIED VALUE
Manual scripting/note-taking/synthesis reduced to review + final decision
Note: Figures reflect the platform's design targets. The 91.7% turnaround reduction is calculated from 3 business days (24 working hours) to 2 hours; remaining ROI metrics should be validated from production measurements after go-live.
Strategic Business Impact
- ✓Accelerated hiring velocity through automated interview prep and rapid evidence consolidation
- ✓Improved candidate experience via adaptive, conversational interviews that respond to real-time signals
- ✓Reduced recruiter cognitive load by automating scripting, note-taking, and assessment synthesis
- ✓Consistent, defensible hiring decisions backed by evidence, rubrics, and AI reasoning trails
- ✓Enterprise integration without disruption to existing recruiter portals, mobile apps, video, and ATS workflows
- ✓Scalable, resilient platform that serves multi-region hiring with business-continuity guarantees